Results 1 to 9 of 9

Thread: Few questions

  1. #1
    Senior Member engracio's Avatar
    Join Date
    Jun 2004
    Location
    Illinois
    Posts
    237

    Few questions

    Hey guys I guess I am back for a while as we wrap up Five or Bust. I've read thru the forum trying to catch up on things. It seems that most are using their quads with half worlers/cpu and two threads per It seem to be the better choice in completing the wu the fastest.

    I know we went with PG last year and it seems to be going fine. With newer and better Prime95 going online where do we stand on second pass and do we still have the option to run it if we chooses to. My thing is that with Prime95 going online a couple years back I know it will catch bad results as we run double check. I am more concern with the legacy client and wu not having second/third matching residue. "It" just seating there waiting for us to say DUH!!. The auto P1 is a great time saver for the project.

    Am I correct that SOB have a certain slice of first pass wu and PG another slice as we move forward? I hope we find another prime soon, it's been a while.

    Thanks

    e

  2. #2
    Hey engracio! Welcome back to the original "Or Bust" project. We could definitely use your prime finding luck on this project! :P

    Pretty much whatever worked well for you with Five or Bust will work well here. Typically I do half the workers and do two threads per worker (or 4 if hyper-threaded) just like you said.

    Currently all second-pass work is being done by the legacy SoB client. I am not sure what you mean with the legacy client not having second/third matching residues, as that is what the legacy client is doing. All results are eventually double-checked. The difference between PG and us is that we have a large gap between when first-pass and second-pass tests are done (second-pass work is currently back in the 13M range) while PG does their double-checking right away. (PG credit is not given until matching residues are returned, so there's a problem with people having to sometimes wait weeks or months to get credit)

    As far as I know there is no way to do second-pass work with Prime95 unless the queue is switched server-side. I don't know what the exact ranges are, but we are in the 19.4M range and PG has been working somewhere in the 18M's, and has recently jumped to 20M so I believe we have the rest of the 19M's.

  3. #3
    Senior Member engracio's Avatar
    Join Date
    Jun 2004
    Location
    Illinois
    Posts
    237
    Thanks enderak, I think you answered all of my questions. I did not know that the legacy client server is still handing out second pass wu. That was my concern, that we have some wu that did not have second/third matching residue yet and we are forging ahead without them.

    I assume prime95 will be able to do second pass once all legacy wu with legacy client are completed with matching residues. That will be in due time.

    btw, try to skip prime95 26.5 build 1 cuz it seems to redo the completed P-1 if you restart your client. George stated he got it fixed on 26.5 build 3

  4. #4
    I thought the residues of Prime95 and the legacy client are compatible, but I may be wrong. So there's no reason the legacy client can't check Prime95's work or vice-versa. I think the main reason was to just keep the old computers busy with useful work until they die off or are otherwise retired. There was some discussion back when the switch was made but I don't remember the details.

    And yes, I noticed that problem with 26.5 and meant to post a message on the Mersenne forum but forgot. Glad to hear it was fixed!

  5. #5
    Senior Member engracio's Avatar
    Join Date
    Jun 2004
    Location
    Illinois
    Posts
    237
    Well maybe the guys "in the know" will swing by and pipe in.

  6. #6
    Congrats for this achievement, with some luck. This last k (40291) has been investigated as the sole candidate from exponents around 5.2M; the odds at finding a PRP were at 50% for 63M; for 9.1M (actual finding) the probability was around 15%. But, back from the beginning of the Five or Bust project, according to the Helm et al. 2008 article in Integers, the odds at finding a PRP for each of the 5 remaining candidate were: 10% for 100M, 50% for 11G and 90% for... 72T!!!. The probability for finding all PRP's at or less than 9.1M was around... 0.9% !!! This results from a sequence of lucky outcomes, of course. The previous one had been the discovery for k=41693 at 5.15M, event which probability (given the state of the art following the discovery for k=2131 at 4.58M) was estimated to ca. 6% according to Phil Moore.

    For the future about dual Sierpinski search:
    - for 78557 < k < 100000, only 2 candidates remain: 79309 (also a direct prime Sierpinski candidate) and 81919, with no PRP for n < 400000
    - for k < 100000 there remain some quasi-Sierpinski dual candidates, i.e. with no prime/PRP, except for very small n (such that 2^n < or ~ k), for n < 400000:
    90527 (prime for only n=1) ; 56839 and 63859 (prime for only n=2) ; 32899 and 55849 (n=10) ; 85489 (n=14) ; 383 (n=15) ; 24737 (n=17) ; 61969 (n=18).
    - for the mixed Sierpinski problem in extended form (neither prime/PRP for k*2^n+1 nor 2^n+k for 78557 < k < 271129), only 2 candidates remain: 79309 (cited above) and 225931 (also a PSP candidate, no PRP for dual, though n has been explored only to 200000).

    Cheers to all.

  7. #7
    Quote Originally Posted by enderak View Post
    I thought the residues of Prime95 and the legacy client are compatible, but I may be wrong. So there's no reason the legacy client can't check Prime95's work or vice-versa. I think the main reason was to just keep the old computers busy with useful work until they die off or are otherwise retired. There was some discussion back when the switch was made but I don't remember the details.
    I happened to have a log file from the old client so I ran the test on Prime95, and the residues do not match so you may be correct engracio. I still recall something about being able to check between them but I can't find it if it was posted on the forum. We'll just have to wait for Louie or someone to chime in on this one I think.

  8. #8
    Senior Member engracio's Avatar
    Join Date
    Jun 2004
    Location
    Illinois
    Posts
    237
    Another question is with almost 700 wu on second pass, is there enough cruncher to complete the legacy client test?

  9. #9
    Member
    Join Date
    Dec 2002
    Location
    Eugene, Oregon
    Posts
    79
    Just responding to Zuzu's thread: I redid the calculations from our INTEGERS paper and find that we had only a 0.55% chance of solving the dual Sierpinski problem by n=9092392, so we were incredibly lucky! The best news is, this is an excellent indication that this problem will also have a successful resolution. I will be crunching here once I get everything finished off on the Five or Busted project, but I think Engracio will bring us luck!

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •